Betting has no downside — make double-or-nothing a real gamble (#950)

[mholzi]

Problem

Reported in discussion #950. The double-or-nothing bet currently has zero downside, so a rational player always bets.

scoring.py:apply_bet_multiplier:

• bet + round_score > 0 → score × 2 ("won")
• bet + round_score == 0 → 0 ("lost")
• no bet → unchanged

A round that scores 0 already breaks the streak and already banks 0 points — so a "lost" bet costs nothing you wouldn't have lost anyway. It's "double or same", not "double or nothing". The bet is free upside and the mechanic is meaningless.

Fix — Design A ("Exact or nothing")

A bet should risk the points you'd otherwise have banked:

• A bet wins only on the exact correct year → round score × 3 (was × 2; bumped because the win condition is stricter).
• A bet that is not exact → 0 for the round, even if the guess would otherwise have scored on a ±tolerance band. You forfeit those points. bet_outcome = "lost".
• A lost bet scores 0 → the streak breaks (unchanged, but now a real consequence).
• No negative scores — the stake is the forfeited round points, the game stays positive-sum.

This makes betting a genuine confidence call: only bet when you're sure you nailed the exact year.

Implementation notes

• apply_bet_multiplier needs to know whether the guess was exact, not just round_score > 0 — pass an is_exact flag (or the year delta) from the caller.
• Update the player-facing bet hint / any TTS or docs copy: "Bet: nail the exact year for 3× — miss it and you score nothing that round."
• Update scoring unit tests.

Closes #950 (tracked from discussion #950).